xref: /freebsd/sys/kern/kern_malloc.c (revision a70cba95822f662d3f9da5119b6a0c433e8f70af)
1 /*-
2  * Copyright (c) 1987, 1991, 1993
3  *	The Regents of the University of California.
4  * Copyright (c) 2005-2009 Robert N. M. Watson
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 4. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)kern_malloc.c	8.3 (Berkeley) 1/4/94
32  */
33 
34 /*
35  * Kernel malloc(9) implementation -- general purpose kernel memory allocator
36  * based on memory types.  Back end is implemented using the UMA(9) zone
37  * allocator.  A set of fixed-size buckets are used for smaller allocations,
38  * and a special UMA allocation interface is used for larger allocations.
39  * Callers declare memory types, and statistics are maintained independently
40  * for each memory type.  Statistics are maintained per-CPU for performance
41  * reasons.  See malloc(9) and comments in malloc.h for a detailed
42  * description.
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_ddb.h"
49 #include "opt_vm.h"
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/kdb.h>
54 #include <sys/kernel.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/mutex.h>
58 #include <sys/vmmeter.h>
59 #include <sys/proc.h>
60 #include <sys/sbuf.h>
61 #include <sys/sysctl.h>
62 #include <sys/time.h>
63 #include <sys/vmem.h>
64 
65 #include <vm/vm.h>
66 #include <vm/pmap.h>
67 #include <vm/vm_pageout.h>
68 #include <vm/vm_param.h>
69 #include <vm/vm_kern.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_page.h>
73 #include <vm/uma.h>
74 #include <vm/uma_int.h>
75 #include <vm/uma_dbg.h>
76 
77 #ifdef DEBUG_MEMGUARD
78 #include <vm/memguard.h>
79 #endif
80 #ifdef DEBUG_REDZONE
81 #include <vm/redzone.h>
82 #endif
83 
84 #if defined(INVARIANTS) && defined(__i386__)
85 #include <machine/cpu.h>
86 #endif
87 
88 #include <ddb/ddb.h>
89 
90 #ifdef KDTRACE_HOOKS
91 #include <sys/dtrace_bsd.h>
92 
93 dtrace_malloc_probe_func_t	dtrace_malloc_probe;
94 #endif
95 
96 /*
97  * When realloc() is called, if the new size is sufficiently smaller than
98  * the old size, realloc() will allocate a new, smaller block to avoid
99  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
100  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
101  */
102 #ifndef REALLOC_FRACTION
103 #define	REALLOC_FRACTION	1	/* new block if <= half the size */
104 #endif
105 
106 /*
107  * Centrally define some common malloc types.
108  */
109 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
110 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
111 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
112 
113 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
114 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
115 
116 static struct malloc_type *kmemstatistics;
117 static int kmemcount;
118 
119 #define KMEM_ZSHIFT	4
120 #define KMEM_ZBASE	16
121 #define KMEM_ZMASK	(KMEM_ZBASE - 1)
122 
123 #define KMEM_ZMAX	65536
124 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
125 static uint8_t kmemsize[KMEM_ZSIZE + 1];
126 
127 #ifndef MALLOC_DEBUG_MAXZONES
128 #define	MALLOC_DEBUG_MAXZONES	1
129 #endif
130 static int numzones = MALLOC_DEBUG_MAXZONES;
131 
132 /*
133  * Small malloc(9) memory allocations are allocated from a set of UMA buckets
134  * of various sizes.
135  *
136  * XXX: The comment here used to read "These won't be powers of two for
137  * long."  It's possible that a significant amount of wasted memory could be
138  * recovered by tuning the sizes of these buckets.
139  */
140 struct {
141 	int kz_size;
142 	char *kz_name;
143 	uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
144 } kmemzones[] = {
145 	{16, "16", },
146 	{32, "32", },
147 	{64, "64", },
148 	{128, "128", },
149 	{256, "256", },
150 	{512, "512", },
151 	{1024, "1024", },
152 	{2048, "2048", },
153 	{4096, "4096", },
154 	{8192, "8192", },
155 	{16384, "16384", },
156 	{32768, "32768", },
157 	{65536, "65536", },
158 	{0, NULL},
159 };
160 
161 /*
162  * Zone to allocate malloc type descriptions from.  For ABI reasons, memory
163  * types are described by a data structure passed by the declaring code, but
164  * the malloc(9) implementation has its own data structure describing the
165  * type and statistics.  This permits the malloc(9)-internal data structures
166  * to be modified without breaking binary-compiled kernel modules that
167  * declare malloc types.
168  */
169 static uma_zone_t mt_zone;
170 
171 u_long vm_kmem_size;
172 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
173     "Size of kernel memory");
174 
175 static u_long kmem_zmax = KMEM_ZMAX;
176 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0,
177     "Maximum allocation size that malloc(9) would use UMA as backend");
178 
179 static u_long vm_kmem_size_min;
180 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
181     "Minimum size of kernel memory");
182 
183 static u_long vm_kmem_size_max;
184 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
185     "Maximum size of kernel memory");
186 
187 static u_int vm_kmem_size_scale;
188 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
189     "Scale factor for kernel memory size");
190 
191 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
192 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
193     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
194     sysctl_kmem_map_size, "LU", "Current kmem allocation size");
195 
196 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
197 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
198     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
199     sysctl_kmem_map_free, "LU", "Free space in kmem");
200 
201 /*
202  * The malloc_mtx protects the kmemstatistics linked list.
203  */
204 struct mtx malloc_mtx;
205 
206 #ifdef MALLOC_PROFILE
207 uint64_t krequests[KMEM_ZSIZE + 1];
208 
209 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
210 #endif
211 
212 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
213 
214 /*
215  * time_uptime of the last malloc(9) failure (induced or real).
216  */
217 static time_t t_malloc_fail;
218 
219 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
220 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
221     "Kernel malloc debugging options");
222 #endif
223 
224 /*
225  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
226  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
227  */
228 #ifdef MALLOC_MAKE_FAILURES
229 static int malloc_failure_rate;
230 static int malloc_nowait_count;
231 static int malloc_failure_count;
232 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN,
233     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
234 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
235     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
236 #endif
237 
238 static int
239 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
240 {
241 	u_long size;
242 
243 	size = vmem_size(kmem_arena, VMEM_ALLOC);
244 	return (sysctl_handle_long(oidp, &size, 0, req));
245 }
246 
247 static int
248 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
249 {
250 	u_long size;
251 
252 	size = vmem_size(kmem_arena, VMEM_FREE);
253 	return (sysctl_handle_long(oidp, &size, 0, req));
254 }
255 
256 /*
257  * malloc(9) uma zone separation -- sub-page buffer overruns in one
258  * malloc type will affect only a subset of other malloc types.
259  */
260 #if MALLOC_DEBUG_MAXZONES > 1
261 static void
262 tunable_set_numzones(void)
263 {
264 
265 	TUNABLE_INT_FETCH("debug.malloc.numzones",
266 	    &numzones);
267 
268 	/* Sanity check the number of malloc uma zones. */
269 	if (numzones <= 0)
270 		numzones = 1;
271 	if (numzones > MALLOC_DEBUG_MAXZONES)
272 		numzones = MALLOC_DEBUG_MAXZONES;
273 }
274 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
275 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
276     &numzones, 0, "Number of malloc uma subzones");
277 
278 /*
279  * Any number that changes regularly is an okay choice for the
280  * offset.  Build numbers are pretty good of you have them.
281  */
282 static u_int zone_offset = __FreeBSD_version;
283 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
284 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
285     &zone_offset, 0, "Separate malloc types by examining the "
286     "Nth character in the malloc type short description.");
287 
288 static u_int
289 mtp_get_subzone(const char *desc)
290 {
291 	size_t len;
292 	u_int val;
293 
294 	if (desc == NULL || (len = strlen(desc)) == 0)
295 		return (0);
296 	val = desc[zone_offset % len];
297 	return (val % numzones);
298 }
299 #elif MALLOC_DEBUG_MAXZONES == 0
300 #error "MALLOC_DEBUG_MAXZONES must be positive."
301 #else
302 static inline u_int
303 mtp_get_subzone(const char *desc)
304 {
305 
306 	return (0);
307 }
308 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
309 
310 int
311 malloc_last_fail(void)
312 {
313 
314 	return (time_uptime - t_malloc_fail);
315 }
316 
317 /*
318  * An allocation has succeeded -- update malloc type statistics for the
319  * amount of bucket size.  Occurs within a critical section so that the
320  * thread isn't preempted and doesn't migrate while updating per-PCU
321  * statistics.
322  */
323 static void
324 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
325     int zindx)
326 {
327 	struct malloc_type_internal *mtip;
328 	struct malloc_type_stats *mtsp;
329 
330 	critical_enter();
331 	mtip = mtp->ks_handle;
332 	mtsp = &mtip->mti_stats[curcpu];
333 	if (size > 0) {
334 		mtsp->mts_memalloced += size;
335 		mtsp->mts_numallocs++;
336 	}
337 	if (zindx != -1)
338 		mtsp->mts_size |= 1 << zindx;
339 
340 #ifdef KDTRACE_HOOKS
341 	if (dtrace_malloc_probe != NULL) {
342 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
343 		if (probe_id != 0)
344 			(dtrace_malloc_probe)(probe_id,
345 			    (uintptr_t) mtp, (uintptr_t) mtip,
346 			    (uintptr_t) mtsp, size, zindx);
347 	}
348 #endif
349 
350 	critical_exit();
351 }
352 
353 void
354 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
355 {
356 
357 	if (size > 0)
358 		malloc_type_zone_allocated(mtp, size, -1);
359 }
360 
361 /*
362  * A free operation has occurred -- update malloc type statistics for the
363  * amount of the bucket size.  Occurs within a critical section so that the
364  * thread isn't preempted and doesn't migrate while updating per-CPU
365  * statistics.
366  */
367 void
368 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
369 {
370 	struct malloc_type_internal *mtip;
371 	struct malloc_type_stats *mtsp;
372 
373 	critical_enter();
374 	mtip = mtp->ks_handle;
375 	mtsp = &mtip->mti_stats[curcpu];
376 	mtsp->mts_memfreed += size;
377 	mtsp->mts_numfrees++;
378 
379 #ifdef KDTRACE_HOOKS
380 	if (dtrace_malloc_probe != NULL) {
381 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
382 		if (probe_id != 0)
383 			(dtrace_malloc_probe)(probe_id,
384 			    (uintptr_t) mtp, (uintptr_t) mtip,
385 			    (uintptr_t) mtsp, size, 0);
386 	}
387 #endif
388 
389 	critical_exit();
390 }
391 
392 /*
393  *	contigmalloc:
394  *
395  *	Allocate a block of physically contiguous memory.
396  *
397  *	If M_NOWAIT is set, this routine will not block and return NULL if
398  *	the allocation fails.
399  */
400 void *
401 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
402     vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
403     vm_paddr_t boundary)
404 {
405 	void *ret;
406 
407 	ret = (void *)kmem_alloc_contig(kernel_arena, size, flags, low, high,
408 	    alignment, boundary, VM_MEMATTR_DEFAULT);
409 	if (ret != NULL)
410 		malloc_type_allocated(type, round_page(size));
411 	return (ret);
412 }
413 
414 /*
415  *	contigfree:
416  *
417  *	Free a block of memory allocated by contigmalloc.
418  *
419  *	This routine may not block.
420  */
421 void
422 contigfree(void *addr, unsigned long size, struct malloc_type *type)
423 {
424 
425 	kmem_free(kernel_arena, (vm_offset_t)addr, size);
426 	malloc_type_freed(type, round_page(size));
427 }
428 
429 /*
430  *	malloc:
431  *
432  *	Allocate a block of memory.
433  *
434  *	If M_NOWAIT is set, this routine will not block and return NULL if
435  *	the allocation fails.
436  */
437 void *
438 malloc(unsigned long size, struct malloc_type *mtp, int flags)
439 {
440 	int indx;
441 	struct malloc_type_internal *mtip;
442 	caddr_t va;
443 	uma_zone_t zone;
444 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
445 	unsigned long osize = size;
446 #endif
447 
448 #ifdef INVARIANTS
449 	KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
450 	/*
451 	 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
452 	 */
453 	indx = flags & (M_WAITOK | M_NOWAIT);
454 	if (indx != M_NOWAIT && indx != M_WAITOK) {
455 		static	struct timeval lasterr;
456 		static	int curerr, once;
457 		if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
458 			printf("Bad malloc flags: %x\n", indx);
459 			kdb_backtrace();
460 			flags |= M_WAITOK;
461 			once++;
462 		}
463 	}
464 #endif
465 #ifdef MALLOC_MAKE_FAILURES
466 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
467 		atomic_add_int(&malloc_nowait_count, 1);
468 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
469 			atomic_add_int(&malloc_failure_count, 1);
470 			t_malloc_fail = time_uptime;
471 			return (NULL);
472 		}
473 	}
474 #endif
475 	if (flags & M_WAITOK)
476 		KASSERT(curthread->td_intr_nesting_level == 0,
477 		   ("malloc(M_WAITOK) in interrupt context"));
478 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
479 	    ("malloc: called with spinlock or critical section held"));
480 
481 #ifdef DEBUG_MEMGUARD
482 	if (memguard_cmp_mtp(mtp, size)) {
483 		va = memguard_alloc(size, flags);
484 		if (va != NULL)
485 			return (va);
486 		/* This is unfortunate but should not be fatal. */
487 	}
488 #endif
489 
490 #ifdef DEBUG_REDZONE
491 	size = redzone_size_ntor(size);
492 #endif
493 
494 	if (size <= kmem_zmax) {
495 		mtip = mtp->ks_handle;
496 		if (size & KMEM_ZMASK)
497 			size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
498 		indx = kmemsize[size >> KMEM_ZSHIFT];
499 		KASSERT(mtip->mti_zone < numzones,
500 		    ("mti_zone %u out of range %d",
501 		    mtip->mti_zone, numzones));
502 		zone = kmemzones[indx].kz_zone[mtip->mti_zone];
503 #ifdef MALLOC_PROFILE
504 		krequests[size >> KMEM_ZSHIFT]++;
505 #endif
506 		va = uma_zalloc(zone, flags);
507 		if (va != NULL)
508 			size = zone->uz_size;
509 		malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
510 	} else {
511 		size = roundup(size, PAGE_SIZE);
512 		zone = NULL;
513 		va = uma_large_malloc(size, flags);
514 		malloc_type_allocated(mtp, va == NULL ? 0 : size);
515 	}
516 	if (flags & M_WAITOK)
517 		KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
518 	else if (va == NULL)
519 		t_malloc_fail = time_uptime;
520 #ifdef DIAGNOSTIC
521 	if (va != NULL && !(flags & M_ZERO)) {
522 		memset(va, 0x70, osize);
523 	}
524 #endif
525 #ifdef DEBUG_REDZONE
526 	if (va != NULL)
527 		va = redzone_setup(va, osize);
528 #endif
529 	return ((void *) va);
530 }
531 
532 /*
533  *	free:
534  *
535  *	Free a block of memory allocated by malloc.
536  *
537  *	This routine may not block.
538  */
539 void
540 free(void *addr, struct malloc_type *mtp)
541 {
542 	uma_slab_t slab;
543 	u_long size;
544 
545 	KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic"));
546 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
547 	    ("free: called with spinlock or critical section held"));
548 
549 	/* free(NULL, ...) does nothing */
550 	if (addr == NULL)
551 		return;
552 
553 #ifdef DEBUG_MEMGUARD
554 	if (is_memguard_addr(addr)) {
555 		memguard_free(addr);
556 		return;
557 	}
558 #endif
559 
560 #ifdef DEBUG_REDZONE
561 	redzone_check(addr);
562 	addr = redzone_addr_ntor(addr);
563 #endif
564 
565 	slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
566 
567 	if (slab == NULL)
568 		panic("free: address %p(%p) has not been allocated.\n",
569 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
570 
571 	if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
572 #ifdef INVARIANTS
573 		struct malloc_type **mtpp = addr;
574 #endif
575 		size = slab->us_keg->uk_size;
576 #ifdef INVARIANTS
577 		/*
578 		 * Cache a pointer to the malloc_type that most recently freed
579 		 * this memory here.  This way we know who is most likely to
580 		 * have stepped on it later.
581 		 *
582 		 * This code assumes that size is a multiple of 8 bytes for
583 		 * 64 bit machines
584 		 */
585 		mtpp = (struct malloc_type **)
586 		    ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
587 		mtpp += (size - sizeof(struct malloc_type *)) /
588 		    sizeof(struct malloc_type *);
589 		*mtpp = mtp;
590 #endif
591 		uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
592 	} else {
593 		size = slab->us_size;
594 		uma_large_free(slab);
595 	}
596 	malloc_type_freed(mtp, size);
597 }
598 
599 /*
600  *	realloc: change the size of a memory block
601  */
602 void *
603 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
604 {
605 	uma_slab_t slab;
606 	unsigned long alloc;
607 	void *newaddr;
608 
609 	KASSERT(mtp->ks_magic == M_MAGIC,
610 	    ("realloc: bad malloc type magic"));
611 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
612 	    ("realloc: called with spinlock or critical section held"));
613 
614 	/* realloc(NULL, ...) is equivalent to malloc(...) */
615 	if (addr == NULL)
616 		return (malloc(size, mtp, flags));
617 
618 	/*
619 	 * XXX: Should report free of old memory and alloc of new memory to
620 	 * per-CPU stats.
621 	 */
622 
623 #ifdef DEBUG_MEMGUARD
624 	if (is_memguard_addr(addr))
625 		return (memguard_realloc(addr, size, mtp, flags));
626 #endif
627 
628 #ifdef DEBUG_REDZONE
629 	slab = NULL;
630 	alloc = redzone_get_size(addr);
631 #else
632 	slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
633 
634 	/* Sanity check */
635 	KASSERT(slab != NULL,
636 	    ("realloc: address %p out of range", (void *)addr));
637 
638 	/* Get the size of the original block */
639 	if (!(slab->us_flags & UMA_SLAB_MALLOC))
640 		alloc = slab->us_keg->uk_size;
641 	else
642 		alloc = slab->us_size;
643 
644 	/* Reuse the original block if appropriate */
645 	if (size <= alloc
646 	    && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
647 		return (addr);
648 #endif /* !DEBUG_REDZONE */
649 
650 	/* Allocate a new, bigger (or smaller) block */
651 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
652 		return (NULL);
653 
654 	/* Copy over original contents */
655 	bcopy(addr, newaddr, min(size, alloc));
656 	free(addr, mtp);
657 	return (newaddr);
658 }
659 
660 /*
661  *	reallocf: same as realloc() but free memory on failure.
662  */
663 void *
664 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
665 {
666 	void *mem;
667 
668 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
669 		free(addr, mtp);
670 	return (mem);
671 }
672 
673 /*
674  * Wake the uma reclamation pagedaemon thread when we exhaust KVA.  It
675  * will call the lowmem handler and uma_reclaim() callbacks in a
676  * context that is safe.
677  */
678 static void
679 kmem_reclaim(vmem_t *vm, int flags)
680 {
681 
682 	uma_reclaim_wakeup();
683 	pagedaemon_wakeup();
684 }
685 
686 #ifndef __sparc64__
687 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
688 #endif
689 
690 /*
691  * Initialize the kernel memory (kmem) arena.
692  */
693 void
694 kmeminit(void)
695 {
696 	u_long mem_size;
697 	u_long tmp;
698 
699 #ifdef VM_KMEM_SIZE
700 	if (vm_kmem_size == 0)
701 		vm_kmem_size = VM_KMEM_SIZE;
702 #endif
703 #ifdef VM_KMEM_SIZE_MIN
704 	if (vm_kmem_size_min == 0)
705 		vm_kmem_size_min = VM_KMEM_SIZE_MIN;
706 #endif
707 #ifdef VM_KMEM_SIZE_MAX
708 	if (vm_kmem_size_max == 0)
709 		vm_kmem_size_max = VM_KMEM_SIZE_MAX;
710 #endif
711 	/*
712 	 * Calculate the amount of kernel virtual address (KVA) space that is
713 	 * preallocated to the kmem arena.  In order to support a wide range
714 	 * of machines, it is a function of the physical memory size,
715 	 * specifically,
716 	 *
717 	 *	min(max(physical memory size / VM_KMEM_SIZE_SCALE,
718 	 *	    VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
719 	 *
720 	 * Every architecture must define an integral value for
721 	 * VM_KMEM_SIZE_SCALE.  However, the definitions of VM_KMEM_SIZE_MIN
722 	 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
723 	 * ceiling on this preallocation, are optional.  Typically,
724 	 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
725 	 * a given architecture.
726 	 */
727 	mem_size = vm_cnt.v_page_count;
728 	if (mem_size <= 32768) /* delphij XXX 128MB */
729 		kmem_zmax = PAGE_SIZE;
730 
731 	if (vm_kmem_size_scale < 1)
732 		vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
733 
734 	/*
735 	 * Check if we should use defaults for the "vm_kmem_size"
736 	 * variable:
737 	 */
738 	if (vm_kmem_size == 0) {
739 		vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
740 
741 		if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
742 			vm_kmem_size = vm_kmem_size_min;
743 		if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
744 			vm_kmem_size = vm_kmem_size_max;
745 	}
746 
747 	/*
748 	 * The amount of KVA space that is preallocated to the
749 	 * kmem arena can be set statically at compile-time or manually
750 	 * through the kernel environment.  However, it is still limited to
751 	 * twice the physical memory size, which has been sufficient to handle
752 	 * the most severe cases of external fragmentation in the kmem arena.
753 	 */
754 	if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
755 		vm_kmem_size = 2 * mem_size * PAGE_SIZE;
756 
757 	vm_kmem_size = round_page(vm_kmem_size);
758 #ifdef DEBUG_MEMGUARD
759 	tmp = memguard_fudge(vm_kmem_size, kernel_map);
760 #else
761 	tmp = vm_kmem_size;
762 #endif
763 	vmem_init(kmem_arena, "kmem arena", kva_alloc(tmp), tmp, PAGE_SIZE,
764 	    0, 0);
765 	vmem_set_reclaim(kmem_arena, kmem_reclaim);
766 
767 #ifdef DEBUG_MEMGUARD
768 	/*
769 	 * Initialize MemGuard if support compiled in.  MemGuard is a
770 	 * replacement allocator used for detecting tamper-after-free
771 	 * scenarios as they occur.  It is only used for debugging.
772 	 */
773 	memguard_init(kmem_arena);
774 #endif
775 }
776 
777 /*
778  * Initialize the kernel memory allocator
779  */
780 /* ARGSUSED*/
781 static void
782 mallocinit(void *dummy)
783 {
784 	int i;
785 	uint8_t indx;
786 
787 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
788 
789 	kmeminit();
790 
791 	uma_startup2();
792 
793 	if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
794 		kmem_zmax = KMEM_ZMAX;
795 
796 	mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
797 #ifdef INVARIANTS
798 	    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
799 #else
800 	    NULL, NULL, NULL, NULL,
801 #endif
802 	    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
803 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
804 		int size = kmemzones[indx].kz_size;
805 		char *name = kmemzones[indx].kz_name;
806 		int subzone;
807 
808 		for (subzone = 0; subzone < numzones; subzone++) {
809 			kmemzones[indx].kz_zone[subzone] =
810 			    uma_zcreate(name, size,
811 #ifdef INVARIANTS
812 			    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
813 #else
814 			    NULL, NULL, NULL, NULL,
815 #endif
816 			    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
817 		}
818 		for (;i <= size; i+= KMEM_ZBASE)
819 			kmemsize[i >> KMEM_ZSHIFT] = indx;
820 
821 	}
822 }
823 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL);
824 
825 void
826 malloc_init(void *data)
827 {
828 	struct malloc_type_internal *mtip;
829 	struct malloc_type *mtp;
830 
831 	KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init"));
832 
833 	mtp = data;
834 	if (mtp->ks_magic != M_MAGIC)
835 		panic("malloc_init: bad malloc type magic");
836 
837 	mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
838 	mtp->ks_handle = mtip;
839 	mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc);
840 
841 	mtx_lock(&malloc_mtx);
842 	mtp->ks_next = kmemstatistics;
843 	kmemstatistics = mtp;
844 	kmemcount++;
845 	mtx_unlock(&malloc_mtx);
846 }
847 
848 void
849 malloc_uninit(void *data)
850 {
851 	struct malloc_type_internal *mtip;
852 	struct malloc_type_stats *mtsp;
853 	struct malloc_type *mtp, *temp;
854 	uma_slab_t slab;
855 	long temp_allocs, temp_bytes;
856 	int i;
857 
858 	mtp = data;
859 	KASSERT(mtp->ks_magic == M_MAGIC,
860 	    ("malloc_uninit: bad malloc type magic"));
861 	KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
862 
863 	mtx_lock(&malloc_mtx);
864 	mtip = mtp->ks_handle;
865 	mtp->ks_handle = NULL;
866 	if (mtp != kmemstatistics) {
867 		for (temp = kmemstatistics; temp != NULL;
868 		    temp = temp->ks_next) {
869 			if (temp->ks_next == mtp) {
870 				temp->ks_next = mtp->ks_next;
871 				break;
872 			}
873 		}
874 		KASSERT(temp,
875 		    ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
876 	} else
877 		kmemstatistics = mtp->ks_next;
878 	kmemcount--;
879 	mtx_unlock(&malloc_mtx);
880 
881 	/*
882 	 * Look for memory leaks.
883 	 */
884 	temp_allocs = temp_bytes = 0;
885 	for (i = 0; i < MAXCPU; i++) {
886 		mtsp = &mtip->mti_stats[i];
887 		temp_allocs += mtsp->mts_numallocs;
888 		temp_allocs -= mtsp->mts_numfrees;
889 		temp_bytes += mtsp->mts_memalloced;
890 		temp_bytes -= mtsp->mts_memfreed;
891 	}
892 	if (temp_allocs > 0 || temp_bytes > 0) {
893 		printf("Warning: memory type %s leaked memory on destroy "
894 		    "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
895 		    temp_allocs, temp_bytes);
896 	}
897 
898 	slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
899 	uma_zfree_arg(mt_zone, mtip, slab);
900 }
901 
902 struct malloc_type *
903 malloc_desc2type(const char *desc)
904 {
905 	struct malloc_type *mtp;
906 
907 	mtx_assert(&malloc_mtx, MA_OWNED);
908 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
909 		if (strcmp(mtp->ks_shortdesc, desc) == 0)
910 			return (mtp);
911 	}
912 	return (NULL);
913 }
914 
915 static int
916 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
917 {
918 	struct malloc_type_stream_header mtsh;
919 	struct malloc_type_internal *mtip;
920 	struct malloc_type_header mth;
921 	struct malloc_type *mtp;
922 	int error, i;
923 	struct sbuf sbuf;
924 
925 	error = sysctl_wire_old_buffer(req, 0);
926 	if (error != 0)
927 		return (error);
928 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
929 	sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
930 	mtx_lock(&malloc_mtx);
931 
932 	/*
933 	 * Insert stream header.
934 	 */
935 	bzero(&mtsh, sizeof(mtsh));
936 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
937 	mtsh.mtsh_maxcpus = MAXCPU;
938 	mtsh.mtsh_count = kmemcount;
939 	(void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
940 
941 	/*
942 	 * Insert alternating sequence of type headers and type statistics.
943 	 */
944 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
945 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
946 
947 		/*
948 		 * Insert type header.
949 		 */
950 		bzero(&mth, sizeof(mth));
951 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
952 		(void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
953 
954 		/*
955 		 * Insert type statistics for each CPU.
956 		 */
957 		for (i = 0; i < MAXCPU; i++) {
958 			(void)sbuf_bcat(&sbuf, &mtip->mti_stats[i],
959 			    sizeof(mtip->mti_stats[i]));
960 		}
961 	}
962 	mtx_unlock(&malloc_mtx);
963 	error = sbuf_finish(&sbuf);
964 	sbuf_delete(&sbuf);
965 	return (error);
966 }
967 
968 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
969     0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
970     "Return malloc types");
971 
972 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
973     "Count of kernel malloc types");
974 
975 void
976 malloc_type_list(malloc_type_list_func_t *func, void *arg)
977 {
978 	struct malloc_type *mtp, **bufmtp;
979 	int count, i;
980 	size_t buflen;
981 
982 	mtx_lock(&malloc_mtx);
983 restart:
984 	mtx_assert(&malloc_mtx, MA_OWNED);
985 	count = kmemcount;
986 	mtx_unlock(&malloc_mtx);
987 
988 	buflen = sizeof(struct malloc_type *) * count;
989 	bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
990 
991 	mtx_lock(&malloc_mtx);
992 
993 	if (count < kmemcount) {
994 		free(bufmtp, M_TEMP);
995 		goto restart;
996 	}
997 
998 	for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
999 		bufmtp[i] = mtp;
1000 
1001 	mtx_unlock(&malloc_mtx);
1002 
1003 	for (i = 0; i < count; i++)
1004 		(func)(bufmtp[i], arg);
1005 
1006 	free(bufmtp, M_TEMP);
1007 }
1008 
1009 #ifdef DDB
1010 DB_SHOW_COMMAND(malloc, db_show_malloc)
1011 {
1012 	struct malloc_type_internal *mtip;
1013 	struct malloc_type *mtp;
1014 	uint64_t allocs, frees;
1015 	uint64_t alloced, freed;
1016 	int i;
1017 
1018 	db_printf("%18s %12s  %12s %12s\n", "Type", "InUse", "MemUse",
1019 	    "Requests");
1020 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1021 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
1022 		allocs = 0;
1023 		frees = 0;
1024 		alloced = 0;
1025 		freed = 0;
1026 		for (i = 0; i < MAXCPU; i++) {
1027 			allocs += mtip->mti_stats[i].mts_numallocs;
1028 			frees += mtip->mti_stats[i].mts_numfrees;
1029 			alloced += mtip->mti_stats[i].mts_memalloced;
1030 			freed += mtip->mti_stats[i].mts_memfreed;
1031 		}
1032 		db_printf("%18s %12ju %12juK %12ju\n",
1033 		    mtp->ks_shortdesc, allocs - frees,
1034 		    (alloced - freed + 1023) / 1024, allocs);
1035 		if (db_pager_quit)
1036 			break;
1037 	}
1038 }
1039 
1040 #if MALLOC_DEBUG_MAXZONES > 1
1041 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1042 {
1043 	struct malloc_type_internal *mtip;
1044 	struct malloc_type *mtp;
1045 	u_int subzone;
1046 
1047 	if (!have_addr) {
1048 		db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1049 		return;
1050 	}
1051 	mtp = (void *)addr;
1052 	if (mtp->ks_magic != M_MAGIC) {
1053 		db_printf("Magic %lx does not match expected %x\n",
1054 		    mtp->ks_magic, M_MAGIC);
1055 		return;
1056 	}
1057 
1058 	mtip = mtp->ks_handle;
1059 	subzone = mtip->mti_zone;
1060 
1061 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1062 		mtip = mtp->ks_handle;
1063 		if (mtip->mti_zone != subzone)
1064 			continue;
1065 		db_printf("%s\n", mtp->ks_shortdesc);
1066 		if (db_pager_quit)
1067 			break;
1068 	}
1069 }
1070 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1071 #endif /* DDB */
1072 
1073 #ifdef MALLOC_PROFILE
1074 
1075 static int
1076 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
1077 {
1078 	struct sbuf sbuf;
1079 	uint64_t count;
1080 	uint64_t waste;
1081 	uint64_t mem;
1082 	int error;
1083 	int rsize;
1084 	int size;
1085 	int i;
1086 
1087 	waste = 0;
1088 	mem = 0;
1089 
1090 	error = sysctl_wire_old_buffer(req, 0);
1091 	if (error != 0)
1092 		return (error);
1093 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1094 	sbuf_printf(&sbuf,
1095 	    "\n  Size                    Requests  Real Size\n");
1096 	for (i = 0; i < KMEM_ZSIZE; i++) {
1097 		size = i << KMEM_ZSHIFT;
1098 		rsize = kmemzones[kmemsize[i]].kz_size;
1099 		count = (long long unsigned)krequests[i];
1100 
1101 		sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
1102 		    (unsigned long long)count, rsize);
1103 
1104 		if ((rsize * count) > (size * count))
1105 			waste += (rsize * count) - (size * count);
1106 		mem += (rsize * count);
1107 	}
1108 	sbuf_printf(&sbuf,
1109 	    "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
1110 	    (unsigned long long)mem, (unsigned long long)waste);
1111 	error = sbuf_finish(&sbuf);
1112 	sbuf_delete(&sbuf);
1113 	return (error);
1114 }
1115 
1116 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
1117     NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
1118 #endif /* MALLOC_PROFILE */
1119